Remote Sensing of Water and Environment

Chapter 4: Optical Remote Sensing of Water and Environment

Ardeshir Ebtehaj
University of Minnesota

4-1 Introduction

Before, we delve into electromegnetic properties and the principles of wave propgation in earth materials, we aim to provide an introduction into remote sensing of some important Earth varaibles related to vegetation, snow, ice and water quality. As previosuly noted, naturally occurring radiation includes reflected solar radiation, which is largely confined to the visible and near-infrared parts of the electromagnetic spectrum (wavelengths between roughly 0.35 and 2.5 ) and thermally emitted radiation. The range of wavelengths generated by a thermally emitting body depends on the temperature. The dominant wavelength is approximately , where T is the absolute temperature of the body. So for a body at a temperature of 273K (0 C), the dominant wavelength is around 11. The distribution of energy over wavelengths falls very sharply for shorter wavelengths, but at longer wavelengths the decrease is much more gradual. For this reason, thermally emitted radiation can be detected both in the thermal infrared part of the electromagnetic spectrum (wavelengths typically 8 to 14 ) and also in the microwave region (wavelengths typically 1 cm to 1 m). The wavelength region between 14 and 1 cm is largely blocked by the atmosphere.
There are several optical and infarred spaceborne instruments notable the Moderate Resolution Imaging Spectroradiometer (MODIS) onboard Aqua and Terra satellites, MultiSpectral Instrument (MSI) onboard Sentinel-2 and LandSat satellites. The MODIS has 36 bands raning from 620 nm to 14.385 with a swath width 2330 km (cross track). The elevation of the satellite is 705 km with 10:30 a.m. descending node (Terra) or 1:30 p.m. ascending node (Aqua). The satellitea are sun-synchronous with near-polar and circular orbits. The resolution are 250 m (bands 1-2), 500 m (bands 3-7), 1000 m (bands 8-36). Visibale to NIR bands (1--19, 600--1000 nm) are used for remote sensing of land, cloud aerosols, ocean colors and atmopsheric water vapor. The bands 20--36 (3.66--14.38 ) are used for remote sensing of surface and cloud tempertuares, cirrus clouds water vapor conetent. ozone and cloud top altitude.
The MSI on Sentinel 2A,B measures the Earth's reflected radiance in 13 spectral bands from VNIR to SWIR (Table 1).
Passive remote sensing systems that detect reflected solar radiation are designed to measure the radiance, i.e., the amount of radiation reaching the sensor in a particular waveband. If the amount of radiation that is incident on the Earth’s surface is known, the reflectance of the surface can be calculated (this requires that the effects of the atmosphere should be corrected).

4-2 Electromagnetic Properties of Snow in the Optical and Near-Infrared Regions

Fresh dry snow appears white to the human eye. That is to say, it is highly reflective, with little variation over the range of wavelengths approximately from 0.4 to 0.65 in which the eye is sensitive. The reason for this lies in the dielectric properties of ice, and the fact that the ice constituting snow is in a very highly divided form.
Reflection coefficient from an ice–air interface and absorption length in ice for electromagnetic radiation with wavelengths between 0.4 and 4.0 (Warren 1984). The absorption length is defined as the distance through which the radiation must travel in order for its intensity to be reduced by a factor of e as a result of absorption alone.
The absorption length of visible-wavelength radiation in ice is of the order of 10 m, which means that in traversing a snowpack with a thickness of (say) 2 m, containing a total thickness of perhaps 1m of ice, a photon has a negligible chance of being absorbed. On the other hand, the photon will encounter of a few thousand air–ice and ice–air interfaces as it traverses the pack, with a probability of about 0.02 of being reflected at each of these interfaces. Thus, it is almost certain that the photon will be scattered back out of the snowpack, and since neither the absorption nor reflection properties of ice vary significantly over the visible waveband, this will be equally true for all wavelengths.
This simple argument implies that the reflection coefficient of a snowpack should be smaller if the grain size is larger, since the number of air–ice interfaces and hence scattering opportunities will be reduced. Furthermore, the generally increasing absorption (smaller absorption lengths) at longer wavelengths implies a corresponding reduction in reflectance at these wavelengths. These phenomena are illustrated in the following figure, which shows the comparative insensitivity of reflectance to grain size in the visible region and the high degree of sensitivity in the range from about 1.0 to 1.3 .
Spectral reflectance of a deep snow pack as a function of grain size (based on Choudhury and Chang 1979).
The reflectance of snow does not depend directly on density, although the processes that cause the increase in density over time also lead to an increase in grain size and hence a consequent decrease in reflectance. As a snow pack ages, it may also acquire a covering of dust or soot that may also decrease the reflectance. While the albedo of a fresh snow cover can exceed 90%, this figure can fall to 40% or even as low as 20% for dirty snow (Hall and Martinec 1985).
The presence of liquid water in a snowpack has little direct effect on its reflectance. The amount of water rarely exceeds 10% by volume and there is in any case sufficient dielectric contrast between water and ice to ensure that the multiple-scattering phenomenon continues to occur. The absorption of electromagnetic radiation in water is similar to that in ice in the visible and near-infrared regions. On the other hand, the presence of liquid water does have an indirect effect on the optical properties, since it promotes clustering of the ice crystals leading to a larger effective grain size and hence lower reflectance. The most significant effect of increasing water content appears to be a small shift of the absorption feature at 1030nm to shorter wavelengths.
Snow is not highly reflective in the thermal infrared region. Figure blow illustrates the variation in reflectance with wavelength from 2 to 14 and with grain size from 10 to 400 . The figure shows that for grain sizes above 100 the reflectance does not exceed 1% throughout the thermal infrared region. The emissivity of dry snow in the thermal infrared region ranges from typically 0.965 to 0.995. In this region of the electromagnetic spectrum the absorption of ice is high, with a maximum near 10 , and the finely divided structure of snow also increases its tendency to act like a black body (i.e., for the emissivity to tend toward 1).
Spectral reflectance of snow in the thermal infrared region. (Original data from Salisbury, D’Aria, and Wald (1994) reproduced by Kuittinen (1997). Reprinted with permission from European Association of Remote Sensing Laboratories.

4-3 Remote Sensing of Snow Cover

Often simple transformation of a few channels or bands will lead to the extraction of useful information from the observed data. One of the simplest types of commonly used band transformations forms a ratio image from two bands. This is often useful when the difference between the pixel values in the two bands is diagnostic of some physical quantity or variable to be investigated, but the pixel values also vary because of, for example, variations in the illumination geometry.

4-3-1 Basic Principles

The snow cover is a large-scale phenomenon. The maximum winter extent in the northern hemisphere is about 46 million square kilometers, but the spatial density of the in situ measurement network is too low to provide an adequate characterization of its distribution (Mognard 2003). In the visible–near-infrared region, snow has a very high albedo. In the microwave region, snow has a brightness temperature that is significantly lower than that of snow-free ground, because of volume scattering in the snowpack.
Snow was observed in the first image obtained from the TIROS-1 satellite in 1960 (Singer and Popham 1963). Snow cover has been routinely monitored from space using optical imagery since 1966 (Matson, Ropelewski, and Varnardore 1986), and since 1978 using passive microwave imagery (Hall et al. 2002). The two approaches have somewhat complementary advantages and disadvantages. While the optical systems (Frei and Robinson 1999) can achieve much higher spatial resolutions, they are limited to daylight and cloud-free conditions. This is a particular difficulty in regions with maritime climates, such as the United Kingdom, where cloud cover is frequently associated with snow cover (Archer et al. 1994). Passive microwave systems can be used at night and through cloud, but their spatial resolution is coarse. These resolutions are generally too coarse for hydrological applications.
A problem common to all approaches to remote sensing of snow cover is that much of the world’s seasonal snow cover occurs in forested areas (Hall et al. 2001) and complex landscapes generally (Walker and Goodison 1993; Solberg et al. 1997; Vikhamar and Solberg 2000; Klein, Hall, and Riggs 1998; Sta¨ hli, Schaper, and Papritz 2002). Therefore, remote sensing capabilities need to have adequate penetration depth through vegetation canopy.
At the finest spatial scales, typical of the spatial resolution of sensors such as Landsat (of the order of a few tens of meters), most approaches to the monitoring of snow cover are based on the analysis of VIR imagery. As pointed out, snow has a very high reflectance, at least for wavelengths below about 0.8 , so it is comparatively easy to detect against a background of snow-free terrain. At the most basic level, single-band imagery can simply be thresholded to discriminate between snow-free and snow-covered terrain.
However, the choice of a suitable threshold value is not always straightforward, and ancillary information may be needed. The threshold may depend on the local imaging geometry, since surfaces that face squarely toward the incident solar radiation will reflect more light to the sensor than those that are oriented more obliquely. A further disadvantage of using single-band imagery is the potential for confusion introduced by cloud cover.
Spectral reflectance of optically thick cloud compared with snow cover of two different grain sizes.
A commoner approach is to use multispectral imagery. Most usefully, this consists of a spectral band in the visible part of the spectrum, plus one centered near 1.65 in the near infrared. The reason for this is that, while the spectral reflectances of snow cover and of cloud are very similar at wavelengths below about 1 (Massom 1991), they diverge in the near infrared and achieve a maximum difference (in the sense that cloud is more reflective than snow) at wavelengths between about 1.55 and 1.75 (Figure above).
For the discrimination of snow, the usual index is called the normalized difference snow index (NDSI), defined in terms of the spectral bands of Landsat TM and Enhanced Thematic Mapper Plus (ETM+) as
where and are the reflectances in bands 2 and 5 (center wavelengths 0.57 and 1.65 ), respectively (Dozier 1984, 1989). Snow is normally assumed to be present if the NDSI exceeds a value of 0.4 (Dozier 1989; Hall, Riggs, and Salomonson 1995), although recent work has suggested that the optimum value of the threshold varies seasonally. Based on field investigations around Abisko, Sweden, Vogel (2002) has suggested that a threshold of 0.48 is more appropriate in July, 0.6 in September.
Snow absorbs in the SWIR, but reflects in the VIR, whereas cloud is generally reflective in these Bands
where cenetral wavelengths at and are and .

4-3-2 Sentinel-2 Algorithm

The following algorithm is for Level-1C products (TOA reflectance) resampled at a spatial resolution of 60 m.

Step 1a: Brightness Thresholds on Red (Band 4):

Step 1b: Normalised Difference Snow Index (NDSI)

Most potential cloudy pixels have NDSI values in a range between -0.1 and +0.2, therefore:

Step 2a: Snow Filter 1: Normalised Difference Snow Index (NDSI)

Cloud and snow reflectances are similar in band 3; however, band 11 reflectance for clouds is very high while it is low for snow.
Pixels with a snow probability higher than zero pass to next step.

Step 2b: Snow Filter 2: Band 8 thresholds

This step eliminate pixels that have high NDSI values and low band 8 (NIR) reflectance.
Pixels with a snow probability higher than zero pass to next step.

Step 2c: Snow Filter 3: Band 2 thresholds

This step eliminates pixels that have high NDSI values and low band 2 (blue) reflectance.
Pixels with a snow probability higher than zero pass to next step.

Step 2d: Snow Filter 4: Ratio Band 2 / Band 4

This step eliminates pixels that have high NDSI values and low Band 2/Band 4 reflectance ratio which usually corresponds to water bodies.
Pixels with a snow probability higher than zero pass to next step.

Step 2e: Processing of Snow Boundaries Zones

This process removes false cloud detection at the boundaries of a snowing region where mixed pixel (snow and ground) could be detected as 'cloud' in the cloud detection algorithm. The 'snow' region is dilated and band 12 reflectance threshold filtering is applied on the dilated boundary zone. Pixels with band 12 reflectance lower than 0.12 have snow confidence that remains unchanged. Pixels that exceed the threshold have their snow probability set to 0.0 and are identified as 'no snow'.

4-4 Remote Sensing of Vegetation Phenology

4-4-1 Basic Principles

For example, healthy vegetation with high concentration of chlorophyll content in thier leaves, reflects more near-infrared (NIR) and green light compared to other VIS wavelengths. But it absorbs more in red and blue range of the spectrum. In general, if there is much more reflected radiation in near-infrared wavelengths than in visible wavelengths, then the vegetation in that pixel is likely to be dense and may contain some type of forest. If there is very little difference in the intensity of visible and near-infrared wavelengths reflected, then the vegetation is probably sparse and may consist of grassland, tundra, or desert. The result of this observation is formulated as the Normalized Difference Vegetation Index (NDVI):
For example, using MODIS data, Red = band 1 (620 – 670 nm) and NIR = band 2 (841 – 876 nm) can be used for NDVI calculation. In general, NDVI values range from -1.0 to 1.0, with negative values indicating clouds and water, positive values near zero indicating bare soil, and higher positive values of NDVI ranging from sparse vegetation (0.1 - 0.5) to dense green vegetation (0.6 and above).
Note: Typically we define near infrared (NIR) from 780 nm to 1400 nm and shortwave infrared (SWIR) from 1400 nm to 3000 nm.
The more a plant is absorbing visible sunlight (during the growing season), the more it is photosynthesizing and the more it is being productive. Conversely, the less sunlight the plant absorbs, the less it is photosynthesizing, and the less it is being productive. Either scenario results in an NDVI value that, over time, can be averaged to establish the "normal" growing conditions for the vegetation in a given region for a given time of the year. In short, a region’s absorption and reflection of photosynthetically active radiation over a given period of time can be used to characterize the health of the vegetation there, relative to the norm.
The difference between the average NDVI for a particular month of a given year (such as August 1993, above) and the average NDVI for the same month over the last 20 years is called NDVI anomaly. (Compare the August 1993 NDVI anomaly to August 1993 NDVI and Average August NDVI in North America.) In most climates, vegetation growth is limited by water so the relative density of vegetation is a good indicator of agricultural drought. The above image shows the NDVI anomaly in the U.S. for August 1993. In that year, heavy rain in the Northern Great Plains (North and South Dakota, Alberta, and Saskatchewan) led to flooding in the Missouri River. The resulting exceptionally lush vegetation appears as a positive anomaly (green). Concurrently, in the Eastern U.S., rainfall was very low, and the region exhibited a strong negative anomaly (dark red).

4-4-2 Sentinel-2 Algorithm

The NDVI ratio for SENTINEL-2 is outlined as follows
where cenetral wavelengths at and are 0.665 and 0.842, respectively.

Step 1:

Step 2:

This process eliminates highly reflective senescing vegetation using filtering tests on band 8/band 3 reflectance ratio. This test is based on the fact that vegetation is highly reflective in NIR and even more reflective in the green. The band 8/band 3 reflectance ratio is higher for vegetation than for cloud or other scene features.
Pixels with a cloud probability higher than zero pass to the next step.

Step 3: Pass 1 for soils detection

The band 2/band 11 ratio values are lower for soils than other scene features including clouds. This identification step is only applied on pixels with band 2 values lower than a threshold that varies linearly as a function of the band 2/band 11 ratio.

Step 4: Pass for water bodies detection

This step identifies different types of water bodies.
The band 2/band 11 ratio values are higher for water bodies than other scene features including clouds.

4-5 Remote Sensing of Water Bodies

Note that Normalized Difference Water Index (NDWI) may refer to one of at least two remote sensing-derived indexes related to liquid water:

4-5-1 Monitoring changes in water content of leaves

We often use NIR and SWIR bands proposed by Gao in 1996:
.
Sentinel-2 NDWI for agricultural monitoring of drought and irrigation management can be constructed using either combinations:

4-5-2 Monitoring water bodies

We often use green and NIR wavelengths, defined by McFeeters (1996):
.
Sentinel-2 NDWI for waterbody detection can be constructed by using:
Note that more recently (https://eoscience.esa.int/landtraining2017/files/posters/MILCZAREK.pdf) a new water mask is introdcued as
.
This index take values from 0 to 12 and an optimal threshold is around 1.4--1.6, above which the water bodies are detected with a high probability.
Sentinel 2A,B: Level 2 Cloud/Snow Detection Algorithm Sequence.

References

  1. Remote Sensing of Snow and Ice, Gareth Rees, ISBN 9780367392307, Published October 2, 2019 by CRC Press.
  2. Web and Sentinel Online materials.